A GMDH-type neural network with multi-filter feature selection for the prediction of transition temperatures of bent-core liquid crystals

Abstract

The QSPR study on transition temperatures of five-ring bent-core LCs was performed using GMDH-type neural networks. A novel multi-filter approach, which combines chi square ranking, v-WSH and GMDH algorithm was used for the selection of descriptors.

Machine learning methods for liquid crystal research: phases, textures, defects and physical properties

“…In two studies, J. Anastasijevic ´and D. Anastasijevic ´predicted the transition temperatures of 243 bent-core liquid crystals using DT, multivariate adaptive regression spine, and a neural network. 33,34 In total, the best performance was obtained with the GMDH-neural network (Group Method of Data Handling) analyzing both two-dimensional (2D) and threedimensional (3D) molecular descriptors optimized by molecular mechanics. The GMDH-MM model utilizes 13 descriptors from 10 different groups, three of them are 3D descriptors including the gravitational index, which gives the atomic masses and their distribution in a molecule and reflects molecular size-dependent bulk effects on the boiling points.…”

Machine learning for soft and liquid molecular materials

“…The quantitative structure–property relationship (QSPR) methodology, combined with a specific type of feed-forward artificial neural network, has been applied to predict the liquid crystallinity and phase transition temperature of bent-core molecules. 48 (Fig. 6a) The authors turned to nonlinear QSPR models and for the first time used a group method of data handling type neural network, testing several machine learning models with different sets of molecular structure descriptors.…”